wrlib/x86_specific.c

   1 /* x86_convert.c - convert RImage to XImage with x86 optimizations
   2  *
   3  *  Raster graphics library
   4  *
   5  *  Copyright (c) 2000 Alfredo K. Kojima
   6  *
   7  *  This library is free software; you can redistribute it and/or
   8  *  modify it under the terms of the GNU Library General Public
   9  *  License as published by the Free Software Foundation; either
  10  *  version 2 of the License, or (at your option) any later version.
  11  *
  12  *  This library is distributed in the hope that it will be useful,
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  *  Library General Public License for more details.
  16  *
  17  *  You should have received a copy of the GNU Library General Public
  18  *  License along with this library; if not, write to the Free
  19  *  Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #include <config.h>
  23
  24 #ifdef ASM_X86
  25
  26
  27 #ifdef ASM_X86_MMX
  28
  29 int
  30 x86_check_mmx()
  31 {
  32     static int result = -1;
  33
  34     if (result >= 0)
  35         return result;
  36
  37     result = 0;
  38
  39     asm volatile
  40         ("pushal                \n" // please don't forget this in any asm
  41          "pushfl                \n" // check whether cpuid supported
  42          "pop %%eax             \n"
  43          "movl %%eax, %%ebx     \n"
  44          "xorl $(1<<21), %%eax  \n"
  45          "pushl %%eax           \n"
  46          "popfl                 \n"
  47          "pushfl                \n"
  48          "popl %%eax            \n"
  49          "xorl %%ebx, %%eax     \n"
  50          "andl $(1<<21), %%eax  \n"
  51          "jz .NotPentium        \n"
  52          "xorl %%eax, %%eax     \n" // no eax effect because of the movl below
  53                                     // except reseting flags. is it needed?
  54          "movl $1, %%eax        \n"
  55          "cpuid                 \n"
  56          "test $(1<<23), %%edx  \n"
  57          "jz .NotMMX            \n"
  58
  59          "popal                 \n" // popal needed because the address of
  60          "movl $1, %0           \n" // variable %0 may be kept in a register
  61          "jmp .noPop            \n"
  62
  63          ".NotMMX:              \n"
  64          ".NotPentium:          \n"
  65          "popal                 \n"
  66          ".noPop:               \n"
  67
  68          : "=m" (result));
  69
  70     return result;
  71 }
  72
  73
  74 /*
  75  * TODO:
  76  *              32/8    24/8    32/16   24/16   32/24   24/24
  77  * PPlain       YES     YES
  78  * MMX                          DONE
  79  *
  80  *
  81  * - try to align stack (local variable space) into quadword boundary
  82  */
  83
  84
  85
  86
  87 void
  88 x86_mmx_TrueColor_32_to_16(unsigned char *image, // 8
  89                            unsigned short *ximage, // 12
  90                            short *err, // 16
  91                            short *nerr, // 20
  92                            short *rtable, // 24
  93                            short *gtable, // 28
  94                            short *btable, // 32
  95                            int dr, // 36
  96                            int dg, // 40
  97                            int db, // 44
  98                            unsigned int roffs, // 48
  99                            unsigned int goffs, // 52
 100                            unsigned int boffs, // 56
 101                            int width, // 60
 102                            int height, // 64
 103                            int line_offset) // 68
 104 {
 105     /*
 106      int x; //-4
 107      long long rrggbbaa;// -16
 108      long long pixel; //-24
 109      short *tmp_err; //-32
 110      short *tmp_nerr; //-36
 111      */
 112
 113     asm volatile
 114         (
 115          "subl $128, %esp               \n" // alloc some more stack
 116
 117          "pushal                        \n"
 118
 119          // pack dr, dg and db into mm6
 120          "movl  36(%ebp), %eax          \n"
 121          "movl  40(%ebp), %ebx          \n"
 122          "movw  %ax, -16(%ebp)          \n"
 123
 124          "movw  %bx, -14(%ebp)          \n"
 125          "movl  44(%ebp), %eax          \n"
 126          "movw  $0, -10(%ebp)           \n"
 127          "movw  %ax, -12(%ebp)          \n"
 128
 129          "movq  -16(%ebp), %mm6         \n" // dr dg db 0
 130
 131          // pack 4|4|4|4 into mm7, for shifting (/16)
 132          "movl $0x00040004, -16(%ebp)   \n"
 133          "movl $0x00040004, -12(%ebp)   \n"
 134          "movq -16(%ebp), %mm7          \n"
 135
 136          // store constant values for using with mmx when dithering
 137          "movl $0x00070007, -16(%ebp)   \n"
 138          "movl $0x00070007, -12(%ebp)   \n"
 139          "movq -16(%ebp), %mm5          \n"
 140
 141          "movl $0x00050005, -16(%ebp)   \n"
 142          "movl $0x00050005, -12(%ebp)   \n"
 143          "movq -16(%ebp), %mm4          \n"
 144
 145          "movl $0x00030003, -16(%ebp)   \n"
 146          "movl $0x00030003, -12(%ebp)   \n"
 147          "movq -16(%ebp), %mm3          \n"
 148
 149          // process 1 pixel / cycle, each component treated as 16bit
 150          "movl 8(%ebp), %esi            \n" // esi = image->data
 151
 152 ".LoopYa:                               \n"
 153          "movl 60(%ebp), %eax           \n"
 154          "movl %eax, -4(%ebp)           \n" // x = width
 155
 156          "movl 64(%ebp), %eax           \n"
 157          "decl %eax                     \n" // y--
 158          "movl %eax, 64(%ebp)           \n"
 159          "js .Enda                      \n" // if y < 0, goto end
 160          "andl $1, %eax                 \n"
 161          "jz .LoopY_1a                  \n" // if (y&1) goto LoopY_1
 162
 163 ".LoopY_0a:                             \n"
 164
 165          "movl 16(%ebp), %ebx           \n" // ebx = err
 166          "movl %ebx, -36(%ebp)          \n" // [-36] = err
 167          "movl 20(%ebp), %eax           \n" //
 168          "movl %eax, -32(%ebp)          \n" // [-32] = nerr
 169
 170          "jmp .LoopXa                   \n"
 171
 172 ".LoopY_1a:                             \n"
 173
 174          "movl 20(%ebp), %ebx           \n" // ebx = nerr
 175          "movl %ebx, -36(%ebp)          \n" // [-36] = nerr
 176          "movl 16(%ebp), %eax           \n" //
 177          "movl %eax, -32(%ebp)          \n" // [-32] = eerr
 178
 179          ".align 16                     \n"
 180 ".LoopXa:                               \n"
 181
 182          // calculate errors and pixel components
 183
 184          // depend on ebx, esi, mm6
 185          "movq (%ebx), %mm1             \n" // mm1 = error[0..3]
 186          "punpcklbw (%esi), %mm0        \n" // mm0 = image->data[0..3]
 187          "psrlw $8, %mm0                \n" // fixup mm0
 188          "paddusb %mm1, %mm0            \n" // mm0 = mm0 + mm1 (sat. to 255)
 189          "movq %mm0, -24(%ebp)          \n" // save the pixel
 190
 191          "movzwl -24(%ebp), %ecx        \n" // ecx = pixel.red
 192          "movl 24(%ebp), %edi           \n" // edi = rtable
 193          "leal (%edi, %ecx, 2), %eax    \n" // eax = &rtable[pixel.red]
 194          "movw (%eax), %dx              \n" // dx = rtable[pixel.red]
 195          "movw %dx, -16(%ebp)           \n" // save rr
 196
 197          "movzwl -22(%ebp), %ecx        \n" // ecx = pixel.green
 198          "movl 28(%ebp), %edi           \n" // edi = gtable
 199          "leal (%edi, %ecx, 2), %eax    \n" // eax = &gtable[pixel.green]
 200          "movw (%eax), %dx              \n" // dx = gtable[pixel.green]
 201          "movw %dx, -14(%ebp)           \n" // save gg
 202
 203          "movzwl -20(%ebp), %ecx        \n" // ecx = pixel.blue
 204          "movl 32(%ebp), %edi           \n" // ebx = btable
 205          "leal (%edi, %ecx, 2), %eax    \n" // eax = &btable[pixel.blue]
 206          "movw (%eax), %dx              \n" // dx = btable[pixel.blue]
 207          "movw %dx, -12(%ebp)           \n" // save bb
 208
 209          "movw $0, -10(%ebp)            \n" // save dummy aa
 210
 211          "movq -16(%ebp), %mm1          \n" // load mm1 with rrggbbaa
 212          "pmullw %mm6, %mm1             \n" // mm1 = rr*dr|...
 213          "psubsw %mm1, %mm0             \n" // error = pixel - mm1
 214
 215
 216          // distribute the error
 217
 218          // depend on mm0, mm7, mm3, mm4, mm5
 219
 220          "movl -36(%ebp), %ebx          \n"
 221
 222          "movq %mm0, %mm1               \n"
 223          "pmullw %mm5, %mm1             \n" // mm1 = mm1*7
 224          "psrlw %mm7, %mm1              \n" // mm1 = mm1/16
 225          "paddw 8(%ebx), %mm1           \n"
 226          "movq %mm1, 8(%ebx)            \n" // err[x+1,y] = rer*7/16
 227
 228
 229          "movl -32(%ebp), %ebx          \n"
 230
 231          "movq %mm0, %mm1               \n"
 232          "pmullw %mm4, %mm1             \n" // mm1 = mm1*5
 233          "psrlw %mm7, %mm1              \n" // mm1 = mm1/16
 234          "paddw -8(%ebx), %mm1          \n"
 235          "movq %mm1, -8(%ebx)           \n" // err[x-1,y+1] += rer*3/16
 236
 237          "movq %mm0, %mm1               \n"
 238          "pmullw %mm3, %mm1             \n" // mm1 = mm1*3
 239          "psrlw %mm7, %mm1              \n" // mm1 = mm1/16
 240          "paddw 8(%ebx), %mm1           \n"
 241          "movq %mm1, (%ebx)             \n" // err[x,y+1] += rer*5/16
 242
 243          "psrlw %mm7, %mm0              \n" // mm0 = mm0/16
 244          "movq %mm0, 8(%ebx)            \n" // err[x+1,y+1] = rer/16
 245
 246
 247          // calculate final pixel value and store
 248          "movl 48(%ebp), %ecx           \n"
 249          "movw -16(%ebp), %ax           \n"
 250          "shlw %cl, %ax                 \n" //NP* ax = r<<roffs
 251
 252          "movl 52(%ebp), %ecx           \n"
 253          "movw -14(%ebp), %bx           \n"
 254          "shlw %cl, %bx                 \n" //NP*
 255          "orw %bx, %ax                  \n"
 256
 257          "movl 56(%ebp), %ecx           \n"
 258          "movw -12(%ebp), %bx           \n"
 259          "shlw %cl, %bx                 \n" //NP*
 260          "orw %bx, %ax                  \n"
 261
 262          "movl 12(%ebp), %edx           \n"
 263          "movw %ax, (%edx)              \n"
 264          "addl $2, %edx                 \n" // increment ximage
 265          "movl %edx, 12(%ebp)           \n"
 266
 267          // prepare for next iteration on X
 268
 269          "addl $8, -32(%ebp)            \n" // nerr += 8
 270
 271          "movl -36(%ebp), %ebx          \n"
 272          "addl $8, %ebx                 \n"
 273          "movl %ebx, -36(%ebp)          \n" // ebx = err += 8
 274
 275
 276          // Note: in the last pixel, this would cause an invalid memory access
 277          // because, punpcklbw is used (which reads 8 bytes) and the last
 278          // pixel is only 4 bytes. This is no problem because the image data
 279          // was allocated with extra 4 bytes when created.
 280          "addl $4, %esi                 \n" // image->data += 4
 281
 282
 283          "decl -4(%ebp)                 \n" // x--
 284          "jnz .LoopXa                   \n" // if x>0, goto .LoopX
 285
 286
 287          // depend on edx
 288          "addl 68(%ebp), %edx           \n" // add extra offset to ximage
 289          "movl %edx, 12(%ebp)           \n"
 290
 291
 292          "jmp .LoopYa                   \n"
 293
 294 ".Enda:                                 \n" // THE END
 295
 296          "emms                          \n"
 297
 298          "popal                         \n"
 299          );
 300 }
 301
 302
 303 #endif /* ASM_X86_MMX */
 304
 305
 306
 307 void
 308 x86_PseudoColor_to_8(unsigned char *image, // 8
 309                      unsigned char *ximage, // 12
 310                      char *err, // 16
 311                      char *nerr, // 20
 312                      short *ctable, // 24
 313                      int dr, // 28
 314                      int dg, // 32
 315                      int db, // 36
 316                      unsigned long *pixels, // 40
 317                      int cpc, // 44
 318                      int width, // 48
 319                      int height, // 52
 320                      int bytesPerPixel, // 56
 321                      int line_offset) // 60
 322 {
 323     /*
 324      * int x; -4
 325      * int cpcpc; -8
 326      *
 327      * int rr; -12
 328      * int gg; -16
 329      * int bb; -20
 330      *
 331      * char ndr; -21
 332      * char ndg; -22
 333      * char ndb; -23
 334      *
 335      * char *err; -32
 336      * char *nerr; -36
 337      *
 338      */
 339     asm volatile
 340         (
 341          "subl $128, %esp               \n" // alloc some stack space
 342          "pushal                        \n"
 343
 344          "movl 44(%ebp), %eax           \n"
 345          "mulb 44(%ebp)                 \n"
 346          "movl %eax, -8(%ebp)           \n" // cpcpc = cpc*cpc
 347
 348          // eax will always be <= 0xffff
 349
 350          // process 1 pixel / cycle, each component treated as 16bit
 351          "movl 8(%ebp), %esi            \n" // esi = image->data
 352
 353 ".LoopYb:                               \n"
 354          "movl 48(%ebp), %ecx           \n"
 355          "movl %ecx, -4(%ebp)           \n" // x = width
 356
 357          "movl 52(%ebp), %ecx           \n"
 358          "decl %ecx                     \n" // y--
 359          "movl %ecx, 52(%ebp)           \n"
 360          "js .Endb                      \n" // if y < 0, goto end
 361          "andl $1, %ecx                 \n"
 362          "jz .LoopY_1b                  \n" // if (y&1) goto LoopY_1
 363
 364 ".LoopY_0b:                             \n"
 365
 366          "movl 16(%ebp), %ebx           \n" // ebx = err
 367 //       "movl %ebx, -36(%ebp)          \n" // [-36] = err
 368          "movl 20(%ebp), %ecx           \n" //
 369          "movl %ecx, -32(%ebp)          \n" // [-32] = nerr
 370
 371          "movl $0, (%ecx)               \n" // init error of nerr[0] to 0
 372
 373          "jmp .LoopXb                   \n"
 374
 375 ".LoopY_1b:                             \n"
 376
 377          "movl 20(%ebp), %ebx           \n" // ebx = nerr
 378 //       "movl %ebx, -36(%ebp)          \n" // [-36] = nerr
 379          "movl 16(%ebp), %ecx           \n" //
 380          "movl %ecx, -32(%ebp)          \n" // [-32] = err
 381
 382          "movl $0, (%ecx)               \n" // init error of nerr[0] to 0
 383
 384
 385          ".align 16                     \n"
 386 ".LoopXb:                               \n"
 387
 388
 389          "movl 24(%ebp), %edi           \n" // edi = ctable
 390          "xorl %edx, %edx               \n" // zero the upper word on edx
 391
 392          // RED
 393
 394          // depends on ebx==err, esi==image->data, edi
 395          "movzbw (%esi), %dx            \n" // dx = image->data[0]
 396          "movsbw (%ebx), %ax            \n" // ax = error[0]
 397          "addw %ax, %dx                 \n" // pixel.red = data[0] + error[0]
 398
 399          "testb %dh, %dh                \n" // test if pixel.red < 0 or > 255
 400          "jz .OKRb                      \n" // 0 <= pixel.red <= 255
 401          "js .NEGRb                     \n" // pixel.red < 0
 402          "movw $0xff, %dx               \n" // pixel.red > 255
 403          "jmp .OKRb                     \n"
 404 ".NEGRb:                                \n"
 405          "xorw %dx, %dx                 \n"
 406 ".OKRb:                                 \n"
 407          "leal (%edi, %edx, 2), %ecx    \n" // ecx = &ctable[pixel.red]
 408          "movl (%ecx), %eax             \n" // ax = ctable[pixel.red]
 409          "movw %ax, -12(%ebp)           \n" // save rr
 410
 411          "mulb 28(%ebp)                 \n" // ax = rr*dr
 412          "subw %ax, %dx                 \n" // rer = dx = dx - rr*dr
 413
 414          "movswl %dx, %eax              \n" // save rer
 415
 416          // distribute error
 417          "leal (, %eax, 8), %ecx        \n"
 418          "subw %dx, %cx                 \n" // cx = rer * 7
 419          "sarw $4, %cx                  \n" // cx = rer * 7 / 16
 420          "addb %cl, 4(%ebx)             \n" // err[x+1] += rer * 7 / 16
 421
 422          "movl -32(%ebp), %ecx          \n" // ecx = nerr
 423
 424          "leaw (%eax, %eax, 4), %dx     \n" // dx = rer * 5
 425          "sarw $4, %dx                  \n" // dx = rer * 5 / 16
 426          "addb %dl, (%ecx)              \n" // nerr[x] += rer * 5 / 16
 427
 428          "leaw (%eax, %eax, 2), %dx     \n" // dx = rer * 3
 429          "sarw $4, %dx                  \n" // dx = rer * 3 / 16
 430          "addb %dl, -4(%ecx)            \n" // nerr[x-1] += rer * 3 / 16
 431
 432          "sarw $4, %ax                  \n" // ax = rer / 16
 433          "movb %al, 4(%ecx)             \n" // nerr[x+1] = rer / 16
 434
 435
 436          // GREEN
 437
 438          // depends on ebx, esi, edi
 439          "movzbw 1(%esi), %dx           \n" // dx = image->data[1]
 440          "movsbw 1(%ebx), %ax           \n" // ax = error[1]
 441          "addw %ax, %dx                 \n" // pixel.grn = data[1] + error[1]
 442
 443          "testb %dh, %dh                \n" // test if pixel.grn < 0 or > 255
 444          "jz .OKGb                      \n" // 0 <= pixel.grn <= 255
 445          "js .NEGGb                     \n" // pixel.grn < 0
 446          "movw $0xff, %dx               \n" // pixel.grn > 255
 447          "jmp .OKGb                     \n"
 448 ".NEGGb:                                \n"
 449          "xorw %dx, %dx                 \n"
 450 ".OKGb:                                 \n"
 451          "leal (%edi, %edx, 2), %ecx    \n" // ecx = &ctable[pixel.grn]
 452          "movw (%ecx), %ax              \n" // ax = ctable[pixel.grn]
 453          "movw %ax, -16(%ebp)           \n" // save gg
 454
 455          "mulb 28(%ebp)                 \n" // ax = gg*dg
 456          "subw %ax, %dx                 \n" // ger = dx = dx - gg*dg
 457
 458          "movswl %dx, %eax              \n" // save ger
 459
 460          // distribute error
 461
 462          "leal (, %eax, 8), %ecx        \n"
 463          "subw %dx, %cx                 \n" // cx = ger * 7
 464          "sarw $4, %cx                  \n" // cx = ger * 7 / 16
 465          "addb %cl, 5(%ebx)             \n" // err[x+1] += ger * 7 / 16
 466
 467          "movl -32(%ebp), %ecx          \n" // ecx = nerr
 468
 469          "leaw (%eax, %eax, 4), %dx     \n" // dx = ger * 5
 470          "sarw $4, %dx                  \n" // dx = ger * 5 / 16
 471          "addb %dl, 1(%ecx)             \n" // nerr[x] += ger * 5 / 16
 472
 473          "leaw (%eax, %eax, 2), %dx     \n" // dx = ger * 3
 474          "sarw $4, %dx                  \n" // dx = ger * 3 / 16
 475          "addb %dl, -3(%ecx)            \n" // nerr[x-1] += ger * 3 / 16
 476
 477          "sarw $4, %ax                  \n" // ax = ger / 16
 478          "movb %al, 5(%ecx)             \n" // nerr[x+1] = ger / 16
 479
 480
 481          // BLUE
 482
 483          // depends on ebx, esi
 484          "movzbw 2(%esi), %dx           \n" // dx = image->data[2]
 485          "movsbw 2(%ebx), %ax           \n" // ax = error[2]
 486          "addw %ax, %dx                 \n" // pixel.grn = data[2] + error[2]
 487
 488          "testb %dh, %dh                \n" // test if pixel.blu < 0 or > 255
 489          "jz .OKBb                      \n" // 0 <= pixel.blu <= 255
 490          "js .NEGBb                     \n" // pixel.blu < 0
 491          "movw $0xff, %dx               \n" // pixel.blu > 255
 492          "jmp .OKBb                     \n"
 493 ".NEGBb:                                \n"
 494          "xorw %dx, %dx                 \n"
 495 ".OKBb:                                 \n"
 496          "leal (%edi, %edx, 2), %ecx    \n" // ecx = &ctable[pixel.blu]
 497          "movw (%ecx), %ax              \n" // ax = ctable[pixel.blu]
 498          "movw %ax, -20(%ebp)           \n" // save bb
 499
 500          "mulb 28(%ebp)                 \n" // ax = bb*db
 501          "subw %ax, %dx                 \n" // ber = dx = dx - bb*db
 502          "movswl %dx, %eax              \n" // save ber
 503
 504          // distribute error
 505          "leal (, %eax, 8), %ecx        \n"
 506          "subw %dx, %cx                 \n" // cx = ber * 7
 507          "sarw $4, %cx                  \n" // cx = ber * 7 / 16
 508          "addb %cl, 6(%ebx)             \n" // err[x+1] += ber * 7 / 16
 509
 510          "movl -32(%ebp), %ecx          \n" // ecx = nerr
 511
 512          "leaw (%eax, %eax, 4), %dx     \n" // dx = ber * 5
 513          "sarw $4, %dx                  \n" // dx = ber * 5 / 16
 514          "addb %dl, 2(%ecx)             \n" // nerr[x] += ber * 5 / 16
 515
 516          "leaw (%eax, %eax, 2), %dx     \n" // dx = ber * 3
 517          "sarw $4, %dx                  \n" // dx = ber * 3 / 16
 518          "addb %dl, -4(%ecx)            \n" // nerr[x-1] += ber * 3 / 16
 519
 520          "sarw $4, %ax                  \n" // ax = ber / 16
 521          "movb %al, 6(%ecx)             \n" // nerr[x+1] = ber / 16
 522
 523          "andl $0xffff, %eax            \n"
 524          // depends on eax & 0xffff0000 == 0
 525          // calculate the index of the value of the pixel
 526          "movw -12(%ebp), %ax           \n" // ax = rr
 527          "mulb -8(%ebp)                 \n" // ax = cpcpc*rr
 528          "movw %ax, %cx                 \n"
 529          "movw -16(%ebp), %ax           \n" // ax = gg
 530          "mulb 44(%ebp)                 \n" // ax = cpc*gg
 531          "addw %cx, %ax                 \n" // ax = cpc*gg + cpcpc*rr
 532          "addw -20(%ebp), %ax           \n" // ax = cpcpc*rr + cpc*gg + bb
 533
 534          "movl 40(%ebp), %ecx           \n"
 535          "leal (%ecx, %eax, 4), %edx    \n"
 536          "movb (%edx), %cl              \n" // cl = pixels[ax]
 537
 538          // store the pixel
 539          "movl 12(%ebp), %eax           \n"
 540          "movb %cl, (%eax)              \n" // *ximage = cl
 541          "incl 12(%ebp)                 \n" // ximage++
 542
 543          // prepare for next iteration on X
 544
 545          "addl $4, -32(%ebp)            \n" // nerr += 4
 546          "addl $4, %ebx                 \n" // err += 4
 547
 548          "addl 56(%ebp), %esi           \n" // image->data += bpp
 549
 550          "decl -4(%ebp)                 \n" // x--
 551          "jnz .LoopXb                   \n" // if x>0, goto .LoopX
 552
 553
 554          "movl 60(%ebp), %eax           \n"
 555          "addl %eax, 12(%ebp)           \n" // add extra offset to ximage
 556
 557          "jmp .LoopYb                   \n"
 558
 559 ".Endb:                                 \n"
 560
 561          "emms                          \n"
 562          "popal                         \n"
 563          );
 564
 565
 566 }
 567
 568 #endif /* ASM_X86 */